Comparative analysis of multi-output machine learning models for solar irradiance and wind speed forecasting: A case study in Tamil Nadu, India

Abstract

The growing share of wind and solar energy has created challenges in electrical networks, mainly due to intermittency, fluctuations, and uncertainty. These issues affect power system stability, grid operations, and the balance between supply and demand. To address this, accurate prediction of solar irradiance and wind speed is critical for integrating renewable energy into power systems. In this study, we propose a multi-output machine learning approach to predict both global horizontal irradiance (GHI) and wind speed simultaneously. The study uses historical meteorological data obtained from the National Solar Radiation Database (NSRDB) for Tamil Nadu, India. Six regression algorithms: linear regression, gradient boosting, random Forest, extreme gradient boosting (XGB), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost) are tested under identical conditions. Model hyperparameters were tuned using GridSearchCV and Bayesian optimization to ensure robust performance. Before modeling, a comprehensive statistical analysis, including input feature distribution and correlation analysis, was conducted. Model accuracy was evaluated using RMSE, MAE, and R² metrics on both training and testing datasets. The results showed that ensemble tree-based methods outperformed the baseline linear model. Among them, CatBoost produced the best results for GHI prediction, while random forest delivered the most reliable wind speed forecasts, demonstrating strong predictive capability for renewable energy applications.